Abstract: The urban low-altitude environment is complex, and it is difficult to effectively perceive and position non-cooperative UAVs with a single sensor, but the integration of multiple types of heterogeneous sensors perceives and positioning data, and there are problems such as data clock and sampling frequency desynchronization, cross-source correlation instability, and noise uncertainty. To solve this problem, this paper proposes a Spatio-Temporal Optimization-driven Dynamic Adaptive Fusion (STO-DAF) framework that considers the temporal and spatial relationships inherent in heterogeneous data. Firstly, a Kinematic-Constrained Trajectory Reconstruction (KCTR) mechanism considering the kinematic constraints of UAV is proposed. The abnormal data correction is carried out through the speed constraint, and the interrupted area of the track is continuously processed in combination with the window threshold to enhance the timing coherence of the data. A phased matching method is constructed to achieve accurate spatiotemporal alignment of heterogeneous data. Secondly, a Spatio-Temporal Optimization mechanism (STO) based on confidence level is designed. By identifying the data source, the correlation and grading mechanism of the confidence and accuracy of matching data pairs is designed, and the data pairs are screened and retained based on the confidence level. Finally, a Dynamic Adaptive data Fusion model (DAF) is constructed. The static model, the optimal noise modeling method based on prior knowledge and posterior estimation strategy is designed, and the accuracy and real-time performance characteristics of mathematical smoothing, geometric clustering and state estimation algorithms are analyzed to cope with time-varying interference, and the above modeling and analysis results are integrated to realize the dynamic fusion and output of the track. The results of multi-height flight experiments in real urban scenes show that the KCTR mechanism can effectively filter out outliers based on physical constraints, and the percentage of positioning accuracy can be improved by 56.21%. The STO mechanism improves the uniformity of spatiotemporal distribution of multi-source data, and can effectively enhance the confidence of the data. The DAF model can perform real-time fusion and trajectory output of multi-source heterogeneous sensor data, with an average calculation time of 0.26 seconds and an average positioning error of 11.06m, and has good generalization ability in real urban scenes.

Key words: non-cooperative UAV perception and positioning, multimodal data fusion, urban low-altitude airspace, low-altitude economy, Low-altitude security